Google has rolled out a new framework called the Agent Payments Protocol (AP2). It’s designed to make transactions handled by agents more secure and flexible across different platforms. This open system was built with input from over 60 companies in the payments and tech worlds. AP2 works as an extension of Google’s earlier protocols, aiming to support various payment methods like credit cards, bank transfers, and cryptocurrencies. The key idea behind it is cryptographically signed “mandates” that help establish trust in every transaction.

How AP2 Works and Who’s Involved

In practical terms, users give an Intent Mandate to tell an agent what they want to do. Then they approve a Cart Mandate to confirm purchase details. For tasks that are delegated, a more detailed Intent Mandate allows the agent to act automatically under set conditions. This setup aims to keep transactions secure and easy to audit, even if the user isn’t directly involved at every step. Big names like American Express, Coinbase, Etsy, Mastercard, PayPal, Salesforce, UnionPay, and others are working with Google on this.

This move seems less about launching a product and more about setting the rules for agent-led shopping. By keeping the system open and compatible with many payment types, Google hopes to create a trusted, scalable environment. But experts point out that just because the system is open or supported by many big companies doesn’t mean it will be widely adopted.

Challenges Around Trust, Liability, and Regulation

One of the biggest questions is: if an agent makes a mistake, who is responsible? Is it the business, the merchant, or the issuer? Until laws and contracts clearly define liability, companies might be cautious. For now, they might limit using AP2 to low-risk tasks, like renewing subscriptions or buying everyday items. Larger purchases could still go through traditional methods or established providers.

A standout feature of AP2 is its use of cryptographically signed mandates. These separate the user’s intent, the purchase details, and payment records, creating a clear trail of what was authorized. This aligns well with regulations like SOX and PCI-DSS, which require proof of user approval. In theory, this setup can help prevent fraud by reducing “agent drift” — when agents act beyond their authority.

However, there are concerns. Experts warn about potential gaps in dispute resolution, jurisdiction issues, identity fraud, and risks from adversarial AI. For example, fully automated purchases could face challenges if signatures aren’t recognized across different regions or if credential providers become points of failure. Plus, mandates alone don’t resolve who’s liable if an agent acts outside its scope.

Implementation and Integration Hurdles

For businesses, adopting AP2 isn’t just about flipping a switch. They’ll need to connect it with existing enterprise systems like ERP and procurement workflows. This requires middleware or API layers to translate AP2 messages into actions within those systems. Governance controls, such as identity management and access controls, will also need to be extended to include AI agents operating under AP2.

Experts suggest that CIOs and CISOs should think about how AP2 fits into their broader security and compliance frameworks. They’ll need to consider how to manage sensitive data, keep track of transactions, and ensure that the framework doesn’t add excessive operational complexity. Compatibility with legacy systems, key management, and audit trails will all be important factors in deciding whether to adopt the new protocol.

In the end, AP2 has the potential to boost the role of AI-driven transactions, but its success will depend on how well companies can address the legal, security, and technical challenges involved. Until liability and dispute resolution are clarified, it’s likely that many enterprises will stick to low-risk uses and wait for more regulatory guidance.

Inspired by

• https://www.computerworld.com/article/4058571/google-unveils-payments-protocol-for-ai-agents-with-major-financial-firms.html

Sources

• gmpg.org
• greyhoundresearch.com
• everestgrp.com
• counterpointresearch.com
• podigee.com